Risk Prediction Of Carotid Atherosclerosis Based On Machine Learning Model And Clinical Characteristics Such As TCM Syndrome Elements

Objective: Analyze the clinical characteristics of TCM syndrome elements of carotid atherosclerosis population,screen the best prediction indicators of carotid atherosclerosis,and establish a simple and effective prediction model of carotid atherosclerosis risk,so as to provide basis for the early identification and formulation of screening indicators of carotid atherosclerosis in health examination population.Methods: A retrospective analysis was made on the physical examination population who underwent neck vascular color Doppler ultrasound and other examinations in the health examination center of Fujian Provincial People’s Hospital from 2017 to 2018.870 physical examination reports meeting the inclusion criteria were screened,including 459 in the carotid atherosclerosis group(hereinafter referred to as CAS group)and 411 in the non carotid atherosclerosis group(hereinafter referred to as non CAS group).The standardized indexes were extracted and summarized into Excel data sheet,including general data,laboratory test data,TCM syndrome element data,color Doppler ultrasound description and results of cervical vessels.Statistical methods were used to test the differences between groups.Further construct the limit gradient boost(XGBoost)model with different input variables and the model based on different algorithms.The prediction performance of the model is mainly evaluated by comparing the specificity,sensitivity,F1 value,area under the working characteristic curve(AUROC)and area under the accurate search curve(AUPRC).Results:1.The age,systolic blood pressure and diastolic blood pressure in CAS group were higher than those in Non-CAS group,statistical test of comparison between groups P<0.05,the difference was statistically significant.2.Total cholesterol,low density lipoprotein,apolipoprotein B,fasting blood glucose,glutamyl transpeptidase,creatinine,urea nitrogen,uric acid,creatine kinase,creatine kinase isoenzyme,lactate dehydrogenase,hydroxybutyrate dehydrogenase and monocyte count in CAS group were higher than those in Non-CAS group,while erythrocyte count and platelet hematocrit were lower than those in Non-CAS group,statistical test of comparison between groups P<0.05,the difference was statistically significant.3.In the population of CAS group,the distribution of disease syndrome elements is the most in the kidney,followed by the liver;the distribution of pathogenic syndrome elements was the most in phlegm,followed by dampness and yin deficiency.4.Comparing the models based on different input variables and different algorithms,the evaluation indexes of XGBoost model based on significantly correlated variables alocation re the highest,with specificity of 0.756,sensitivity of 0.804,F1 value of 0.796,AUROC of0.850 and AUPRC of 0.853.For the recognition of CAS,the greatest contribution of clinical features is age,followed by systolic blood pressure,and creatinine;among the syndrome elements,the most contributing factor is the disease location element table,followed by the disease element cold and phlegm.Conclusion:1.The median age was 62 years old,and the normal high blood pressure was 135 / 79 mm Hg,which was related to the occurrence of CAS.2.Higher concentrations of total cholesterol,low density lipoprotein,apolipoprotein B,fasting blood glucose,glutamyltranspeptidase,creatinine,urea nitrogen,uric acid,creatine kinase,creatine kinase isoenzyme,lactate dehydrogenase,hydroxybutyrate dehydrogenase,monocyte count,lower concentrations of erythrocyte count and platelet hematocrit were associated with carotid atherosclerosis.3.The frequency distribution characteristics of TCM syndrome elements in CAS group,the disease location syndrome elements are mainly kidney and liver,and the disease nature syndrome elements are mainly phlegm,dampness and yin deficiency.4.XGBoost model based on significant difference variables has the best effect in identifying CAS,with AUROC of 0.84,which can effectively distinguish whether subjects have CAS or not.In the best XGBoost model,age,systolic blood pressure,creatinine,disease location factor table,disease factor cold and sputum can be used as screening indicators to identify CAS.